You Want It Darker?

Making art with artificial intelligence may draw you to the monstrous.

by

Last August, I ran into Sami, a local DJ, outside the Yellowknife Walmart. Sami is an art angel – stubbornly creating spacey ambient music in a town that much prefers Moose FM’s country-trucker-rock. But that evening, in his white sweatsuit and sky-blue Crocs, his face and beard radiant in the setting sun, Sami looked like Moses down from Sinai. Some change had washed over him. 

“Have you tried Midjourney yet?” he asked, his eyes alight with hope. He was talking about the latest artificial intelligence technology, an art-generator that mines a vast database of images to make original compositions. All users need to do is tell it, via typed command, what they want to see.

I shook my head. I had read articles about it but was wary of so-called proprietary technology. Sami fixed me in his benevolent gaze. His sweatsuit began to glow. His beard turned to tinsel. The hydraulics of Walmart’s automatic doors made a low-fi, rhythmic whoosh-whoosh-whoosh. Here was an emissary from the future. “Don’t be afraid,” he said. “Anything is possible when you speak poetry to robots.” 

I fell hard for the pitch. It was the promise of word power that hooked me, the chance to create with my own utterances, like God in the book of Genesis: “Let there be light!” I wasn’t prepared, though, for just how supernatural it would be.

In 2014, Elon Musk spoke at an MIT symposium. “With artificial intelligence,” he told the crowd, “we are summoning the demon.” My first experiences did feel a little ouija. I tried speaking poetry to a Midjourney bot, typing in T.S. Eliot’s famous line “not with a bang but a whimper.” In 17 seconds, a ghost appeared – the edges of its white drapery soaked in blood, a burning British parliament in the distance. It was a haunting interpretation of Eliot’s poem The Hollow Men. Looking at ‘my’ first AI art sent a tingle up my spine. The bot appeared to know things. 

Artificial intelligence is adept at making apocalyptic, sci-fi imagery. Horror imagery, though, is perhaps what it is best at. Midjourney allows users to scroll through images as they are being made by people around the world. I sat in front of my screen late into the night watching demon after demon appear in hyper-realistic detail – wizened, mottled, malicious. Scarier still were the vacant-eyed beings. I was unnerved. Why do so many users want to conjure nightmares rather than dreams? 

In September, social media was abuzz with news about Loab, a particularly terrifying female demon accidentally summoned by the Swedish artist Steph Maj Swanson, also known as Supercomposite. Ascribing a numerical weight to each word in the command can guide the bot’s priorities. Supercomposite was playing around with negatively weighted text prompts when the ghoul appeared. Negative prompts sometimes send bots looking for imagery that has an abstract relationship to the typed command rather than a literal one, but Loab arrived completely unbidden.

Artificial intelligence commentators talk about latent spaces, the thick darknesses between data, the murky minus-world. Natural language processing is the branch of AI that gives bots the ability to understand the nuances of words that humans do, but it now operates at a level not even its creators fully comprehend. This, Stephen Marche wrote for The Atlantic, is “inherent to the abyss of deep learning.”  

“Loab is an emergent phenomenon that arises in certain AI image synthesis models,” states the Loabmancer website that chronicles her appearances. “She likely lives in the outer reaches of the latent space and can be accessed with negatively weighted prompts. Even more unsettlingly, using images of Loab to query the model often results in disturbing, gruesome imagery – often of dismembered women and children. It is not known exactly why Loab has arisen.”

Perhaps Loab’s Orpheus-like journey is simply a consequence. The data that AI bots mine is beyond-belief huge. DALL-E 2, another text-to-image AI generator, was apparently trained using 650 million image-caption pairs. It has, according to an article in Hyperallergic, seen more images, paintings and other visual phenomena than most human experts. AI then, is the biggest trick mirror in the world, reflecting back everything humanity has already made. It then distorts it, via the human impulse to forgo restraint. We must make it weirder; we must make it darker. With Loab, we got what we asked for. 

AI is prone to creepiness. It defaults to the monstrous. It improvises but cannot yet do so with subtlety. 

As a catergory of AI research, ‘what we ask for’ is perhaps more interesting than the art being made. I’m a writer, an armchair sociologist. I enjoy observing the dissonance between final images and their original text command, more so than looking at AI art without its ‘title.’ What we ask for, it turns out, is drama. Cinematic lighting, high-octane render and volumetric fog are common requests. Occasionally, there’s a food craving like “seamless texture of roasted chicken skin.” One night, a user named Melissa764 asked the bot a question: “How do you think my feelings are?” I could picture her – quiet, mousy, sincere. I could also picture the author of “Baby Jesus Meets an Alien to eat Taco Bell on a Thursday to Discuss Stocks.” A wise-cracking, obnoxious ‘bro,’ obviously. 

Not all AI users are engaged in the dark arts. In his first AI frenzy, Sami produced futuro-medieval castles enveloped in lavender mists. I tend toward speaking fantasy sneakers into existence – apocalyptic orthotics by Adidas. Shoes that look like a cross between Air Jordans and Pinocchio. But I want to focus this essay on darkness, as that is how the technology feels to me. The categories of black and white are too reductive when discussing art, but AI is inherently black.  

AI darkness, fascinatingly, is due in large part to the fact that the technology, at least as of last month, is not fully formed. AI cannot always make an accurate realistic human face. It cannot do fingers well. Nor does it always understand positional terms. It often mixes up ‘on’, ‘in’ and ‘over’, for instance. A simple “woman on couch” command created a woman growing out of a couch, torso fused with cushion. Her head was on backwards. I didn’t ask for a backwards head.

But AI is prone to creepiness. It defaults to the monstrous. It improvises but cannot yet do so with subtlety. Yes, Peter Paul Rubens painted a gruesome and violent Saturn Devouring His Son in the 17th century. But Midjourney went further with my prompt, “Saturn Eating His Children,” offering crows pecking at a writhing mass of putrid entrails.

We are at an interesting time in AI art history – the beginning. Embryonic toddler-bots, still learning to talk, have given us a new digital aesthetic. Flesh bunches up in strange ways. Hands sprout misshapen fingers. Arms appear in threes, or jut at dislocated angles. Let’s call this aesthetic the Uncanny Grotesque. 

The Uncanny Grotesque also smushes materials together in paradoxical ways. DALL-E 2 is the better program for design, and there are accomplished, tech-savvy artists who are making new super-robust sculptural surfaces; chunky silk, puffy granite, rubbery wood. It’s as though AI is adding new elements to the periodic table (“Let there be concrete-jello!”). 

When AI does get faces right, they still don’t look, well, human. In its chilling AI issue earlier this year, Adbusters Magazine illustrated the difference, contrasting art by humans and art by machines. The bot-generated images have consciousness, but no soul. There’s an alien glint in those beautiful, oversized eyes. The magazine defended the human artist, reminding us that feats of AI are trophies for corporations and governments. The caption reads: “We take something that is fundamentally human. We feed it into the mill. We grind it to grist.”

There are artists who employ AI’s Uncanny Grotesque purposefully, to make a point deeper than surface shock (or schlock) value. Thanks to the Internet, we live in a general state of shock-malaise anyway. Even the outraged mobs scream their slogans on autopilot now.  It is no surprise that Montreal artist Jon Rafman, the godfather of post-Internet art, is using AI to make ramped-up, agitated social critique. Meanwhile, Sofia Crespo calls herself a “generative artist entangling with artificial life-forms.” She makes AI images and videos that are eerie and prophetic – goldfish with feathers and six eyes. 

Bots are great at caricature. They are great at exaggeration, at tacky, at mockery. And if AI-generated people don’t look fully human, they can at least tell us something about humanity. Spend time on Supercomposite’s AI Instagram page if you enjoy the merging of repulsive with beautiful, or if you enjoy not enjoying it. Her Sea of Torment could be a contemporized version of Edvard Munch’s The Scream. I find it hard to look at her art. It costs me something to look, some part of myself I’d prefer not to lose. My innocence? My fear of the dark? I’m not quite sure. But it feels spiritually uncomfortable. 

The biggest enemy of impact is volume. There is too much AI art to scroll through. Millions of images – most of them unsatisfying – are generated every day. I sometimes get ‘art sick’ from scrolling on AI channels – where overstimulation manifests as a depressed kind of boredom. Ben Davis, in his recent book Art in the After-Culture, makes this prediction about the rise of AI, “we’ll find ourselves numbed, worn down by the onslaught of novelty.” This has already happened, hasn’t it? Sometimes it feels like there’s so much nothingness on social media, just so much hyper, high-strung, derivative nothing. How much more of the ‘new’ can our exhausted psyches take?  Here’s hoping the spate of articles questioning whether AI art is actually art, will slowly fade away. A better question now is: “What is still meaningful?”

Sami posted a photo on his Instagram account the other day. A frosty forest scene. Subarctic spruce trees in winter. He was taking photos of the real world again. I commented – “Not AI!” He replied “Not AI. Just. Pure. Air.” 

As we start to assimilate AI into everyday art and everyday life, this lack of restraint, this making darker, making weirder – may give us surges of surprise, who knows. Maybe there is something new under the sun after all, an immaculate oddness, something perfectly and glowingly other, that will lead us to marvel and laugh and weep. Will it be made by fully formed sentient intelligences? Be afraid. Anything is possible when you speak poetry to robots. ■

PS: Worried you missed something? See previous Galleries West stories here or sign up for our free biweekly newsletter.